Meta's Fundamental AI Research (FAIR) team announced the release of five new artificial intelligence (AI) research models. These models have various functionalities, such as text and image generation, AI-generated speech detection, and code completion. One of the models, called Chameleon, is capable of understanding and generating both images and text. It can take input that includes text and images and produce a combination of text and images. Meta mentioned that this feature could be used to generate captions for images or create new scenes using text prompts and images. Pretrained models for code completion were also released. These models were trained using Meta's multitoken prediction approach, which involves training large language models (LLMs) to predict multiple future words simultaneously, rather than predicting one word at a time. Another model, JASCO, provides more control over AI music generation.

Instead of relying solely on text inputs, this model can accept various inputs like chords or beats, allowing the incorporation of symbols and audio in a single text-to-music generation model. A model called AudioSeal introduces an audio watermarking technique that enables localized detection of AI-generated speech. It can pinpoint AI-generated segments within larger audio snippets and detect AI-generated speech up to 485 times faster than previous methods. The fifth AI research model released aims to enhance geographical and cultural diversity in text-to-image generation systems. Meta has provided geographic disparities evaluation code and annotations to improve evaluations of text-to-image models. Meta mentioned in an earnings report that it plans to invest $35 billion to $40 billion in AI and metaverse development by the end of 2024. Meta CEO Mark Zuckerberg also highlighted the company's various AI services, including AI assistants, augmented reality apps, and business AIs. To stay updated on AI news, you can subscribe to the daily AI Newsletter from PYMNTS.